| name | chromadb-rag-workflow |
| description | Design or review a ChromaDB-backed RAG workflow. Use when: creating collections, choosing embedding functions, chunking documents, writing retrievers. Do not use: for general LLM integration, non-RAG patterns. |
ChromaDB RAG Workflow
Build Chroma-backed retrieval with stable collections, metadata, and evaluation.
Process
- Identify source documents, update cadence, tenancy/security boundaries, and
retrieval success criteria.
- Choose
PersistentClient or HttpClient based on deployment shape.
- Define collection naming, embedding function/model, document IDs, and
metadata schema.
- Choose chunking and normalization before ingestion.
- Ingest with stable IDs and metadata needed for filters and debugging.
- Add retrieval tests for expected matches, metadata filters, no-match cases,
and stale/deleted document behavior.
- Document backup, rebuild, and re-embedding strategy.
- Keep retrieval metrics or qualitative eval examples close to the feature.
Rules
- Do not change embedding model for an existing collection without a
re-embedding plan.
- Do not delete/recreate persistent collections as normal cleanup.
- Do not rely on vector similarity alone when tenant, document type, or
freshness filters are known.
- Keep source documents recoverable so vector data can be rebuilt.
Output
- Collection and metadata design
- Ingestion/retrieval flow
- Persistence and backup plan
- Retrieval evaluation checklist
